Austerity and tax compliance∗ Francesco Pappad`a†

Yanos Zylberberg‡

September 2, 2017

Abstract Relying on a novel measure of VAT compliance in a panel of 35 countries, we document a robust negative response of tax compliance to changes in tax rates. In order to rationalize this finding, we develop a theoretical framework where heterogeneous firms adjust the share of declared activity. We calibrate the model using firm-level data in Greece, and find large leakages following the recent austerity plans. We then show how differences in financial development and the size of economic activity at the margin of informality are able to explain the heterogeneous response of tax compliance to tax rates across countries. JEL Classification Codes: E02, E62, H26. Key words: austerity, tax compliance, credit frictions.

∗

A previous draft of this paper has circulated under the title “Austerity Plans and Tax Evasion: Theory and Evidence from Greece”, Banque de France Working Paper n. 546. We thank the editor Ayse Imrohoroglu and two anonymous referees for their insightful comments, as well as Manolis Galenianos, Yuriy Gorodnichenko, Francesco Lippi, Claudio Michelacci, Daphne Nikolitsa, Marco Pagano, and seminar participants at the “Macro and Policy Implications of Underground Economy and Tax Evasion” at Bocconi, 2014 Annual Meeting of the Society for Economic Dynamics (SED) Toronto, 15th Journ´ees LAGV Aix-en-Provence, University of Lausanne, Einaudi Institute for Economics and Finance, Italian Treasury, CSEF Napoli and Athens University of Economics and Business. The views expressed in this paper are those of the authors and do not necessarily represent those of the Banque de France or of the Eurosystem. † Corresponding author. Banque de France, International Macroeconomics Division, 31 rue Croix des Petits Champs, 75001 Paris. E-mail: [email protected]. ‡ University of Bristol, Priory Road complex, BS1 8TU Bristol. E-mail: [email protected].

1

1

Introduction

Recent episodes of austerity in peripheral European countries have shown that the increase in tax revenues after a tax hike may be undermined by the strong response of tax compliance. While fluctuations in output due to tax hikes have been studied in the literature,1 the aggregate implications of the response of tax compliance to such reforms have been overlooked. A decrease in tax compliance may not only reduce the impact of a fiscal adjustment, but also have collateral effect on firm decisions, e.g., through their access to external finance and investment. In this paper, we provide evidence of a large and negative tax compliance response to fiscal consolidation episodes. We compile a catalogue of indirect tax reforms in 35 countries between 1990 and 2012, and construct a measure of Value-Added-Taxes (VAT) compliance based on the comparison between two distinct sources: reported household consumption collected through household surveys and tax revenues reflecting taxpayers’ declarations.2 We find a robust and sizeable negative association between changes in VAT compliance and changes in VAT rates. A VAT increase of 10% is associated with a decrease in VAT compliance of about 4%. Importantly, the elasticity of tax revenues to tax reforms is much smaller within the sample of countries with high legal enforcement. In order to explain the quantitative response of tax compliance to tax reforms as well as its heterogeneity across institutional environments, we build a theoretical model with heterogeneous credit-constrained firms operating in a modern or a traditional sector. In the model, there is imperfect tax enforcement. Transparency, that is the share of declared activity, simultaneously determines access to external finance and tax pressure. A tax increase has two distinct effects on transparency. First, firms at the margin of informality do not find profitable anymore to be transparent and get access to external finance. As a consequence, they switch their activity from the modern (and transparent) sector to the traditional (and informal) sector. Second, medium-size firms still operate mainly in the modern sector but they declare less than before. At both ends of the firm size distribution, however, the response to tax increase does not pass through a transparency adjustment: very large firms remain fully transparent while very small firms remain fully informal. We explore the quantitative implications of our model that can be measured 1

See for instance Alesina and Ardagna (2009), Romer and Romer (2010), Favero et al. (2011), Auerbach and Gorodnichenko (2012), Ilzetzki et al. (2013), and Alesina et al. (2015). 2 This measure of VAT compliance is constructed to be orthogonal to fluctuations in each 2-digit category of goods. In an economy where taxpayers declare entirely their transactions, the ratio of actual annual tax revenues to the expected VAT revenues would be equal to 1 independently of fluctuations in sectoral output.

2

against the initial stylized facts. We reproduce the fiscal adjustment program imposed on the Greek government by the “Troika” (EU-ECB-IMF) in 2010. We calibrate the model to match firm-level balance sheets data on 30’000 Greek firms provided by Hellastat. We find that the elasticity of tax revenues to tax hikes is about 0.5 as both transparency and output respond to the tax hikes. Three quarters come from the transparency component, against one quarter explained by the contraction in output. These estimates are consistent with the drop in VAT compliance and tax revenues observed after the fiscal consolidation in Greece.3 We then analyze how (i) fundamentals of the economy – tax enforcement, financial development and firm size distribution – and (ii) characteristics of tax policies affect the magnitude of the transparency response. In a first step, we run a series of counterfactual experiments and show that the aggregate transparency response to the baseline tax policy strongly depends on financial development and firm size distribution. Indeed, the differential response to the tax policy depends on the share of economic activity generated by firms at the margin of informality, which is a function of both fundamentals. For instance, Southern European countries are economies where the aggregate response of tax compliance to tax hikes is large, because tax enforcement is weak and firms at the margin of informality constitute a large share of the economy. In a second step, we modify the characteristics of the baseline tax reform to account for policy features observed in other peripheral European economies. We find that size-dependent tax reforms may be more effective than uniform tax hikes. The rationale for size-dependent tax policies in the model is that the elasticity of tax revenues to tax rates markedly depends on firm size, and is close to zero or even negative for very small firms. The results highlighting the role of fundamentals for the response of tax compliance should be interpreted cautiously. The quantitative analysis relies on a static model in a closed economy with a mass of heterogeneous entrepreneurs, one production factor, one final good and an exogenous government. The supply of entrepreneurs and their endowments is fixed, and the main adjustment variable is transparency which governs both access to credit and tax pressure. As the firm size distribution is constant, the model does not provide insights about the long-term effects of taxation on the structure of production.4 In addition, a proper welfare 3

Following the 2010 tax reforms in Greece, VAT compliance decreased by about 7% while the VAT rate increased by about 20%, that amounts to an elasticity of VAT compliance to VAT rates of -0.35. An implication is that only half of the expected additional tax revenues was collected – 1.5 instead of 3.1 points of GDP – leading to further adjustments in 2011, 2012 and 2013. 4 See Fortin et al. (1997) for a theoretical analysis of the effect of taxation on firm distribution, and (Kumar et al., 1999; Garicano et al., 2016) for an empirical investigation of the determinants of firm size.

3

analysis would need to balance the immediate consumption costs of fiscal consolidation with possible gains in future consumption, and our model does not account for such gains. Our contribution to the economic literature is twofold. First, we build a measure of tax compliance and study its dynamics, in particular its response to tax reforms. Second, we study the aggregate implications of the elasticity of tax compliance to tax reforms in a model with heterogeneous firms and credit frictions. The empirical analysis of fluctuations in tax compliance relates to a public economics literature which analyzes tax evasion using micro-level data.5 Our empirical approach exploits discrepancies between two reporting sources of income, as in Fisman and Wei (2004); Cai and Liu (2009); Kleven et al. (2011) for instance. In contrast with this literature, we provide model-based estimates and study the implications of fluctuating tax compliance on aggregate variables. The existence of a tax compliance response beyond the traditional behavioral response to tax rates is often ignored in the macroeconomic literature. A recent and notable exception is Pappa et al. (2015), which study the contribution of tax evasion and corruption to the size of the fiscal multiplier during the recent consolidation plans in Greece, Italy, Portugal and Spain. In line with our main findings, they show that tax hikes increase the incentives to conceal part of the activity and produce in the less productive informal sector, thus increasing output losses. The paper speaks to a large and mostly theoretical literature examining the relationship between informal activity and financial development. Specifically, we relate to Gordon and Li (2009) in that the interaction of financial development and imperfect tax enforcement shapes the efficiency of tax reforms. Gordon and Li (2009) study the implications of such interaction for optimal tax policy, and rationalize the differences in tax structure between developing and developed countries. A crucial assumption in this literature is that firm transparency affects access to credit, a feature that is also present in Straub (2005) and Desai et al. (2007) for instance. Two recent contributions provide empirical support for this hypothesis and document a robust and positive correlation between credit and firm transparency. Capasso and Jappelli (2013) exploit local variations in the size of the underground economy while Ellul et al. (2015) provide a micro-based analysis relating reliance on external finance to accounting transparency and tax pressure. Finally, our modeling with a dual technology borrows from the literature on shadow economies.6 We slightly depart from this literature (Rauch, 1991; Straub, 5

See Andreoni et al. (1998); Slemrod and Yitzhaki (2002) for a review, and Pomeranz (2015) for an empirical investigation of VAT evasion and third-party reporting. 6 See Enste and Schneider (2000); Porta and Shleifer (2008) for a review.

4

2005) because we allow firms to adjust their degree of informality rather than being fully informal or fully transparent. The paper is organized as follows. In Section 2, we introduce a measure of tax compliance and show its response to fiscal consolidation episodes. Motivated by this empirical analysis, we describe our quantitative framework in Section 3. In Section 4, we calibrate our model using quasi-exhaustive firm-level balance sheet data in Greece. We match important moments of the distribution of firms, and study the evolution of the calibrated economy under the 2010 austerity program. In Section 5, we then generalize our analysis and show the role of fundamentals of our economy, i.e. legal enforcement and the number of firms at the margin of informality, and the characteristics of tax reforms for the aggregate response of tax compliance to tax rates. Finally, Section 6 concludes. 2

The elasticity of tax compliance to fiscal consolidation

In this section, we show that fluctuations in tax compliance should not be overlooked when evaluating the aggregate response of an economy to fiscal consolidation. In this regard, we compile a catalogue of tax reforms, associate indirect tax rates and reported consumption for 2-digit categories of goods, and construct an annual measure of VAT compliance for about 35 countries between 1990 and 2012.7 We then study the dynamics of tax compliance and show that there exists a strong and robust negative association between changes in tax compliance and tax rates in our sample of 35 countries, even when controlling for fluctuations in output. Tax compliance is not only a factor that affects tax revenues in the long run but it also fluctuates markedly. Importantly, the elasticity of tax compliance to tax rates is larger among countries of our sample with lower legal enforcement. 2.1

Data sources

To construct a measure of VAT compliance, we use three different data sources.8 First, we collect a decomposition of tax revenues by tax instruments, as declared by each national tax authority. Second, we collect household consumption measures as 7

A recent contribution constructs a similar measure for EU countries (CPB, 2013) between 2000 and 2011. In contrast with this study, our dataset includes non-EU countries and covers a larger time span. 8 Our measure of tax compliance identifies tax evasion from discrepancies between two reporting sources of income, a strategy similar to Cai and Liu (2009) and Kleven et al. (2011) for instance. Closer to our empirical measure, Fisman and Wei (2004) look at the discrepancies between the declared exports of Hong Kong to China and the imports of China from Hong Kong to measure empirically tax evasion, and they show that these discrepancies are not due to measurement errors as they are systematically higher for those goods subject to higher taxation.

5

reported by national statistical offices and use these 2-digit good category consumption measures as a proxy for the potential tax base. Third, we create a dataset of VAT reforms in which we document changes in tax rates, but also and more crucially changes in the tax base. Tax revenues. For European Union countries, data on tax revenues are collected centrally by Eurostat under the European system of national and regional accounts (ESA 2010) transmission programme. European Union member states are legally obliged to transmit this information.9 VAT revenues are generally recorded on an accrual basis, which may introduce some noise in reports. For instance, revenues are calculated based on the transaction time but there could be a time difference between transaction and cash receipt. Along the same lines, a transaction can be declared but may not be collected. For non-EU/OECD members, we take advantage of the OECD Model Tax Convention to collect revenues from the category Value added type taxes (VAT). Finally, we complete our data on tax revenues by directly collecting detailed national accounts from national statistical offices. Household consumption. In countries complying with the ICLS Resolution on household income and expenditure statistics (2003), household consumption is collected by statistical offices using integrated surveys (e.g., the Income, Expenditure and Wealth questionnaires). One purpose of this data collection is to estimate baskets of goods and services and compute consumer price indices. There are two differences with tax reporting that are worth mentioning. First, while tax revenues are based on taxpayers’ declarations, the Households Final Consumption Expenditure is reported by (final good) buyers who are not legally accountable for tax evasion. Second, household surveys are less standardized across countries than accounting reports and there may exist significant variation in survey design. There is a number of measurement issues arising with these surveys, and the extent to which they affect the final consumption measure may differ across countries. We list several of them below. First, there may exist a difference between the time of acquisition and the time of use. Our ideal measure should be computed using the time of acquisition for a better comparison with VAT revenues. Second, expenditure may be computed as an accounting variable, i.e., a purchase value or a payment approach, or a consumption costs approach estimating the service flow from the good acquisition. Third, goods may be consumed gradually. The choice 9

The format is a list of National taxes (the National Tax List) that are subsequently aggregated in common categories. As regards VAT, we focus on category D.211: Value added type taxes (VAT).

6

between a payment approach and a consumption costs approach is then particularly relevant. In most cases, statistical offices use the payment approach which fits our purposes. Apart from definitions, there are other differences across household surveys that could affect our analysis. Some statistical offices use a fixed reference period and then interview all households during a short period of time, some others define a moving reference period usually corresponding to the 12 previous months. Data collection may also vary: the process could be guided by an interviewer or the questionnaire can be completed by a household member. The scope of the survey and the sampling design are less problematic. Sample size and questions are designed such as to provide precise estimates for the baskets of goods and services, which is also our variable of interest. One difference of our measure with standard price indices is that we do not sum the different consumption categories to determine a cost of living, but we also weight them by their supposed VAT rate to determine the VAT content of the consumption basket. Valued-Added Tax rates. We retrieve VAT rates and we reference the types of goods (at the 2 digit level) that are subject to these rates for each country/year. This data collection supports our empirical approach and allows us to estimate fluctuations in VAT compliance with sufficient precision. For example, in a large number of countries, categories like medical services, international public transport, basic food products or cultural services are subject to reduced rates or exemptions and—very importantly—the set of exempt categories are frequently updated. For European Union members after 2005, we use two administrative sources, the European Commission and Eurostat, to construct the monthly VAT rates for about 80 expenditure categories. We complete this dataset for European Union members before 2005 and other countries by using national sources. This exercise requires some harmonization across expenditure categories in different accounting systems. It also requires some cleaning that we describe in the online Appendix. Finally, we collect other taxes such as to control for simultaneous changes in income tax or corporate tax. Among the sample of 35 countries, we can identify 65 major VAT reforms implying changes in VAT rates larger than 5%. About 45 of these reforms are associated with an increase in the effective VAT rate and can be labelled as episodes of “fiscal consolidation”. In contrast, about 20 reforms imply lower effective rates. These large fiscal episodes are quite equally spread across countries.

7

2.2

A measure of VAT compliance

In order to construct a measure of VAT compliance, we need declared transactions and a counterfactual measure that would capture the actual transactions. We use VAT revenues as a proxy for declared transactions. The difference between declared transactions and VAT-charged transactions is minimal and comes from arrears from defaulting legal entities. We use household-based consumption measures as a proxy for actual transactions based on the fact that households are not liable for reporting consumption of undeclared transactions.10 Aggregate tax compliance is the ratio between tax revenues from total declared transactions and the counterfactual tax revenues from actual transactions. Letting DTt,c,j (resp. ATt,c,j ) denote the declared (resp. actual) transactions and τt,c,j denote the good-specific VAT rates for each good j in country c and year t, our measure T Ct,c,j of VAT compliance is defined as: T Ct,c

P j τt,c,j DTt,c,j . = P j τt,c,j ATt,c,j

Letting Tt,c denote VAT revenues in year t for country c, and Ct,c,j the reported consumption of good j in year t and country c, we have that: T Ct,c = P

Tt,c . j τt,c,j Ct,c,j

Intuitively, the measure T Ct uses two different sources and capture the discrepancies between the two sources. We argue that such discrepancies capture, for a large share, tax evasion decisions. It is true, however, that they also may capture loose enforcement from tax authorities. For instance, tax authorities may tolerate informal exemptions for some sectors, regions, or newly-taxed activities. Given the availability (time coverage and data quality) of our main data sources across countries, our final dataset covers an unbalanced panel of 35 (mostly European) countries between 1990 and 2012.11 There exist several adjustments that we need to implement for our empirical 10

In few countries, however, there may exist an interaction between the type of transactions and household reports. Indeed, in China, consumers have incentives to ask for VAT revenues as they serve for national lottery and households may better keep track of registered transactions than undeclared transactions. In this case, we would underestimate fluctuations in tax compliance. 11 The list of countries in the sample is: Australia, Austria, Belgium, Bulgaria, Canada, Chile, Colombia, Cyprus, Czech Republic, Denmark, Estonia, Finland, France Germany, Greece, Hungary, Ireland, Israel, Italy, Japan, Korea, Luxemburg, Malta, Netherlands, New Zealand, Norway, Poland, Portugal, Roumania, Russia, Serbia, Slovak Republic, Slovenia, South Africa, Spain, Sweden, Switzerland and the United Kingdom.

8

measure to be as close as possible from the theoretical benchmark T Ct,c . First, we are interested in the short-term fluctuations of this measure, and we need to smooth the “high-frequency” measurement error. Tax reforms are often implemented during the year, while national accounts are closed at the end of each period, i.e., year or quarter. For this reason, we need to generate the effective tax rate for a unit of consumption in a given period. When tax rates were changed during the course of the year, we construct the annual effective tax rate by weighting each tax rate by the consumption observed during its spell. When consumption could not be observed at a higher frequency than the period, we construct the annual effective tax rate by weighting each tax rate by the duration within the period during which it was enforced. Second, some tax reforms did not modify rates but also modify the category of goods that are subject to the different tax regimes. In such instances, we redefine our tax base. Third, some reforms modify the tax environment without modifying the tax rates per se. For instance, a reform may considerably simplify the registration process. We collect this information and control for any such reforms in our regression analyses. We describe some of these challenges in greater details in the online Appendix. 2.3

Tax compliance and tax reforms

To uncover the correlation between VAT compliance and VAT reforms, we estimate the following specification in first difference: ∆T Ct,c = αat,c + β∆τt,c + γ · Xt,c + µc + νt + εt,c ,

(1)

where t stands for years and c stands for the country. ∆ is the log-difference operator, i.e., ∆xt = ln(xt ) − ln(xt−1 ), atc is the HP-filtered GDP per capita, and τtc is the effective VAT tax rate. The vector X includes time-varying controls, such as the existence of concomitant tax reforms, growth in trade balance, sectoral output and government expenditures. µc captures the country-specific trends in tax compliance, and νt captures year-Fixed Effects. εt,c is the error term. The coefficient of interest β can be interpreted as the elasticity of tax compliance to tax rates conditional on output fluctuations. We report in Table 1 the results of specification (1) estimated on our sample of countries. We find an elasticity of VAT compliance to standard VAT rate of about −.40 (see panel A) and −.45 with respect to effective VAT rates (see panel B), i.e., when weighting standard, reduced and super-reduced rates by their incidence. In the benchmark regression (column 1), we control for country-specific trends,

9

year-Fixed effects and the country economic cycle. The estimates are extremely robust to the addition of controls for government expenditures (column 2), sectoral composition (column 3) and trade (column 4).12 This estimated elasticity is an average measure across potentially very different countries – with strong or weak institutions. To uncover the differences across countries, we use a simple dichotomy exercise and separate our sample between above- and below-median countries along the Time of debt enforcement proceedings, as measured in Djankov et al. (2008).13 We then estimate the elasticities βh (belowmedian) and βl (above-median) in a similar specification as (1) with interactions. We report the separate estimates in Table 2. While the elasticity of tax compliance to tax rates is around βl = −.50 for low-enforcement countries, it falls around βl = −.20 for high-enforcement countries (see columns 3 and 4). In the remainder of the paper, we develop an analytical framework which helps to understand (i) the magnitude of the empirical elasticity, and (ii) the possible sources of differences across countries. 3

A model of firm transparency and investment

This section presents a static model of firm transparency and investment which allows to derive macro-elasticities of tax revenues, transparency and output to taxes, accounting for firm size heterogeneity. There are two crucial ingredients in our framework, and two main margins of adjustments. We assume that entrepreneurs face financial frictions under the form of a collateral constraint, and there is imperfect tax enforcement. We further assume that one margin of adjustment, transparency, governs both access to credit and tax pressure.14 This assumption relies on the hypothesis that concealed activity is less pledgeable than declared activity. A lower tax compliance would reduce the capacity 12

The estimates are also robust when we control for simultaneous income or corporate tax reforms. 13 Countries below-median are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Japan, Luxembourg, Netherlands, New Zealand, Norway, Sweden, Switzerland, United Kingdom. 14 The perfect correlation between transparency and access to credit seems to contradict recent findings in Artavanis et al. (2016). They find that the ratio of credit over income granted by bankers depends on the income declared by the borrower and the bankers’ beliefs on undeclared activity. Bankers anticipate how reported income maps into real income based on the occupation of borrowers. Occupations characterized by high tax evasion are therefore those which are offered large loans relatively to their reported income. However, this result does not imply that borrowers can pledge their concealed activity exactly as their reported activity. Instead, the amount of credit is a function of their reported income multiplied by an occupation-specific factor, and a higher reported income for a given individual in each occupation would still be associated with looser credit constraints.

10

to levy funds. We also introduce two technologies, a traditional technology, and a more productive modern technology which requires an innovation. The fixed cost to operate in the modern sector implies that very small firms, which are not able to levy sufficient funds for investment in the modern technology to be profitable, mostly operate in the informal sector with the traditional technology and without external financing. For the sake of exposure, the model specification is minimal. We ignore demand effects, and, as there is a unique final good, there is no substitutability between differentiated goods as a response to differential taxation. There is a unique production factor whose supply is infinitely elastic, subject to a collateral constraint. The supply of entrepreneurs and their endowments is considered as given in our closed economy, and we ignore the impact of taxation on the general structure of production including foreign direct investments. The government policy is taken as given and the model is static such that we do not rationalize the use of a fiscal consolidation. We assume that a tax authority exogenously sets an audit schedule. Finally, we disregard fiscal corruption, which is potentially important in countries with relatively low tax enforcement. While corruption does not modify tax receipts for the government for a given transparency, it does modify the entrepreneur’s behavior ex-ante through the share that is captured by corrupt officials. Fiscal corruption could affect our conclusions if the surplus share that is captured by fiscal officials responds to taxes. 3.1

Environment

The economy lasts for one period and is populated by a continuum of risk-neutral entrepreneurs of measure one. Each entrepreneur is endowed with ω. Let G(·) denote the cumulative distribution of endowments.15 Firms produce a unique consumption good and use capital as the unique production factor. The market for the consumption good is perfectly competitive and there is an infinitely elastic demand at price p = 1. There are two technologies available to entrepreneurs in order to produce the consumption good: a traditional one and a modern one. With the modern technology, capital can be used to produce the consumption good according to the following production function: f (k) = Ak α . 15 We use heterogeneity in endowments in order to generate differential response to tax hikes across firms. However, one can think that such heterogeneity can also be indirectly generated by differences in fundamentals, e.g., heterogeneity in productivity, or in industry (and their reliance on external financing).

11

By contrast, the returns on the traditional technology are linear and equal to ρ. The access to the modern technology is conditional on producing an “innovation”, whose success is subject to an idiosyncratic draw depending on innovation efforts. When an entrepreneur invests c, the entrepreneur gets access to the modern technology with probability p(c). Since the access to the modern technology is stochastic, there may exist relatively informal firms using the traditional technology and relatively transparent firms using the modern technology for a same initial endowment. We turn to the firm organization. Each entrepreneur owns a unique firm that is organized in a unit mass of homogeneous plants. The plants or establishments are homogeneous in the sense that entrepreneurs cannot use a different technology or different production factors across plants. We assume however that entrepreneurs can choose the fraction of plants whose value added is concealed. Each plant is either fully declared or fully informal, and let γ denote the fraction of declared plants (thereafter transparency). By assumption, γ is the share of declared collateral and output. There is a tax authority which mechanically raises taxes τ on the reported value added, i.e., the value added generated in the declared establishments.16 The tax authority has access to an audit technology and monitor firms in order to retrieve the concealed value added. Let z(ω) denote the probability to be monitored for a firm with initial endowment ω, and let us assume that z is an increasing function.17 When a concealed plant is detected by the tax authority, firms pay the tax θτ on the concealed value added that is retrieved. θ ≥ 1 is the exogenous punishment for being detected. We turn to financial markets. We assume that the economy is small and that the international financial market is willing and able to supply an unlimited amount of risk-less bonds that yield the international interest rate r > 0. Among entrepreneurs, those with small endowments might want to borrow in order to expand their investment in the modern technology. They can do so by issuing bonds, which are subject to a financial friction. Entrepreneurs can only pledge a share λ of declared endowment to their creditors.18 As a result, entrepreneurs face the following credit 16

We assume that there is only one unique tax rate on a unique final good. Accordingly, our model will not be able to generate the dynamic compositional effects justifying our careful empirical construction of tax compliance (see Section 2). In our quantitative results, we consider as fixed the share of firms subject to reduced and super-reduced tax rates (see Section 4). 17 This assumption can be rationalized with a tax authority optimizing its auditing behavior, and subject to fixed or linear auditing costs as a function of endowment. An increasing auditing schedule would then arise as the equilibrium schedule, as smaller firms have otherwise higher incentives to be transparent (due to the collateral constraint). 18 We assume here that only endowment can be pledged and not future earnings. The reason

12

constraint, which crucially depends on transparency:19 (1 + r)(k − ω) ≤ λγω.

(2)

The timing of actions is as follows. Entrepreneurs invest in innovation, receive the innovation draw and decide whether to adopt the modern technology or not upon innovation success. Further, entrepreneurs decide on their level of transparency, which is going to jointly determine how many plants can be pledged to lenders and taxes to be paid to the government. They borrow capital (k − ω) at the international interest rate subject to the pledgeability constraint. Finally, they produce and reimburse their creditors. In parallel, once a technology has been adopted, the tax authority monitors with an audit effort z and firms pay taxes or fines following the audit outcome. We have not specified yet whether entrepreneurs could have incentives in lending their endowment. We assume (i) that the return to the traditional technology is equal to the international interest rate ρ = r and (ii) that credit is fully transparent and taxed at the same rate τ . This implies that (i) entrepreneurs prefer to invest in the traditional technology rather than lending, except if they are fully transparent, and (ii) never borrow to produce in the traditional technology. In the following lines, we describe the equilibrium allocation characterizing our economy. In order to clarify the entrepreneurs’ trade-off between low tax compliance and access to credit, we start with the entrepreneur’s program once innovations have been made. 3.2

The entrepreneur

The traditional entrepreneur. We first consider an entrepreneur endowed with ω and the traditional technology, subject to an audit effort p(ω) from tax authorities. This traditional entrepreneur solves the following program: πωtr = max {[1 − τ γ − (1 − γ)θz(ω)τ ]rω} γ

is essentially that, when entrepreneurs can freely adjust the extent to which they declare activity, they may not have any commitment to report future earnings once they have received funds from the investors. By contrast, endowment can be pledged when debt is contracted. 19 Note that creditors can only seize a fraction of entrepreneur’s endowment in transparent plants, and taxes are junior to this recovery process. This assumption rules away a potential gamblingfor-resurrection behavior from entrepreneurs where they would evade taxes and leave creditors with the tax arrears in case of monitoring. We assume that creditors observe firm’s endowment and technology but the recovery technology is fully inefficient at recovering funds from concealed plants. Alternatively, we can relax this assumption and assume that there exists two technologies λt > λc to recover funds from transparent and concealed firms.

13

The entrepreneur never borrows nor lends, and invests exactly her endowment. Her transparency choice, however, depends on how θz(ω) compares to 1. Strictly above 1, she becomes fully transparent (γ = 1). Strictly below 1, she remains fully informal (γ = 0). Otherwise, she is indifferent and any γ solves the optimization program. The modern entrepreneur. We now consider an entrepreneur endowed with ω and the modern technology, and subject to an audit effort z(ω) from tax authorities. This modern entrepreneur maximizes her profits net of taxes subject to the credit constraint of equation (2): πωmd = max {[1 − τ γ − (1 − γ)θz(ω)τ ]Ak α − r(k − ω)} , γ,k

subject to (1 + r)(k − ω) ≤ λγω. 1

As long as θp(ω) ≤ 1 and ω ≤ (A(1 − τ )α/r) 1−α , the credit constraint is binding.20 In this case, the solution kˆ verifies: Aαk

α−1



(1 + r)[1 − θz(ω)]τ 1 − θz(ω)τ − λ



 1+αk −1 = r, α ω

(3)

and the transparency choice γˆ is obtained by substituting the solution kˆ into the credit constraint. Equation (3) is very intuitive.21 There is a trade-off between borrowing, and reaping the high returns in the modern technology and the cost that it represents in terms of transparency. In order to borrow an additional unit from lenders, the firm needs to declare part of its activity and pay taxes (second term in the square brackets below). At the optimum, the difference between the gain and the cost should be equal to the price r of borrowing. When the credit constraint is not binding, the solution to the program is close to the solution for the traditional technology case. The entrepreneur invests up to her 1 optimal level (A(1−τ γ −(1−γ)θz(ω)τ )α/r) 1−α and lends the rest of her endowment. The transparency choice depends on how θz(ω) compares to 1. Strictly above 1, she becomes fully transparent. Strictly below 1, she remains fully informal. Otherwise, she is indifferent and any γ solves the optimization program. 20

1

(A(1 − τ )α/r) 1−α is the frictionless optimal level of capital, so that firms with an endowment higher than this level are not financially constrained. 21 It could be that the solution to this equation implies that transparency is greater than 1. In this case,  ˆ k = min{ (λ+1+r)ω , k} 1+r γ = min{1, γˆ }

14

We now need to determine what is the initial choice, i.e., the investment in innovation c. We describe this choice and define the equilibrium of our economy next. 3.3

Equilibrium

Given the audit schedule z(ω), the entrepreneur solves:  max p(c)πωmd (z(ω)) + [1 − p(c)]πωtr (z(ω)) − c , c

which brings:   0 p (c) πωmd (z(ω)) − πωtr (z(ω)) = 1.

(4)

As firm size increases, the innovation cost gets relatively smaller compared to the gains, i.e., the differences between operating with the traditional or modern technologies increase, and firms invest more in innovation. As a result, the share of firms that innovate and use the modern technology increases with firm size. Naturally, since the incentives to innovate are crucially related to the differential gains between the two technologies, any downward shift in the returns to the modern technology, for instance through more stringent credit constraint or higher taxes, will reduce the investment in innovation for all firms. Equation 4 completes our set of equations characterizing the equilibrium: Definition 1. Equilibrium. Each entrepreneur of each type ω chooses the investment in innovation c (equation 4), observes the realization of the investment and produces with the modern or traditional technology, maximizes profits subject to the credit constraint (equation 2), and determines the level of capital and transparency (equation 3). The equilibrium allocation may be described by looking at two distinct endowment regions. When endowment is small, entrepreneurs need access to credit, and transparency and leverage depend on firm endowment through two channels. First, the probability to operate with the modern technology increases with endowment. Second, firms borrow such as to bridge the gap between their wealth and the optimal investment (which should imply that transparency decreases with endowment), but the difference between paying and evading taxes depends on the response of tax authorities. When firms are large enough, they do not borrow anymore and transparency increases with endowment such as to leave the absolute value of concealed production constant.

15

3.4

The behavioral response of the economy to a tax increase

We now illustrate important implications of the model. Consider the comparative statics exercise in which an adverse tax shock affects our economy, i.e., an increase in τ . In our framework, we can distinguish two effects related to such tax increase. In the modern sector, transparency choices, leverage and production depend on the level of taxes. After an increase in taxes, declaring more plants in order to relax the credit constraints is costlieer and entrepreneurs conceal more. This effect can be interpreted as the intensive margin effect, i.e. modern firms adjusting their transparency. In parallel, taxes also depress investment in innovation such that higher taxes induce a lower share of firms operating in the modern sector. This effect can be interpreted as the extensive margin effect. In general, both the intensive and extensive margins work in the same direction and their intensity is mostly concentrated in small and medium size firms relying on external finance. We turn now to the aggregate response. Let εωτγv , εωγ , εωv denote, respectively, the elasticity of tax revenues, transparency and output with respect to taxes for any given endowment ω. dx τ εωx = dτ x We can define equivalent aggregate elasticities as follows: Z |

εωτγv dG(ω) {z

ετ γv

}

Z =1+

εωγ dG(ω) + {z

|

εγ <0

}

Z

εωv dG(ω) {z } | εv <0

Notice that our elasticities are not the elasticities of aggregate quantities with respect to taxes, but rather the individual elasticities with respect to taxes weighted by their prevalence over the population of firms. In practice, our weighted elasticities will be very close to the elasticities of aggregate quantities. This is a small open economy: prices (including bond prices) are fixed such that there are no general equilibrium effects. The aggregate elasticities are thus easy to derive from each entrepreneur’s decisions. Before turning to the quantitative analysis, we also need to define what is the role of transparency in the output drop captured by εv . We decompose the response of output to taxes as follows: εv = νv + νγ . The response of output to taxes measured by the elasticity εv has two components: the direct component νv and the the indirect component νγ . The direct compo16

nent is defined as νv = εv,γ=¯γ , and is the response of output to taxes maintaining transparency fixed. The elasticity νv therefore measures the standard output drop in response to a tax hike, which is due to the lower expected returns in investment. The second component νγ measures the indirect impact of transparency on the output drop. As transparency falls in response to the tax hikes, the firm leverage decreases, which leads to a drop in output. In the following section, we illustrate the quantitative importance of each elasticity, as we calibrate our model to the Greek economy to reproduce the recent austerity plans. 4

Austerity in Greece

We provide in this section a quantitative analysis of the aggregate transparency response to tax hikes. We build on the previous theoretical framework and calibrate the model on a benchmark economy, i.e., Greece just before the 2010 adjustment program. The organization of this section is as follows. We first study the crisis episode through the lens of our model: we provide some numerical estimates for the elasticities of transparency and output. We then provide additional insights on the distributional implications of tax hikes in our framework and discuss their empirical support. 4.1

The benchmark calibration

Our model allows us to match precise moments of firm heterogeneity. Naturally, these degrees of freedom are obtained at the expense of some others: we consider the size distribution of firms as exogenous based on the fact that firm endowment is not as responsive as investment or transparency. Similarly, we shut down the possibility for technology and other fundamentals of the economy to evolve. We calibrate the model using firm-level balance sheet data from Hellastat.22 This dataset consists in comprehensive balance sheet information of Greek firms over the period 2001-2013. Firms have to publish their balance sheets whenever two of the following three criteria are fulfilled: (i) Turnover: 3 Million, (ii) Total Assets: 1.5 Million, (iii) Average staff: 50 people. We therefore observe the universe of registered firms above these thresholds and smaller firms that publish their accounts on a voluntary basis. After cleaning the data for missing observations, we are left 22

We thank the research director of the Foundation for Economic and Industrial Research (IOBE), Aggelos Tsakanikas, and Evaggelia Valavanioti for giving us access to Hellastat data.

17

with more than 25’000 firms per year. The dataset is an unbalanced panel.23 Our sample of firms represents a very high share of Greek economic activity (more than 80%). Firms with assets above 9 Million Euros are observed with certainty and very small firms (with assets below 100,000 Euros) are mostly unobserved. Between those two thresholds, we only observe a sub-sample of firms, which, in practice, may be biased. Figure 2 shows that the firm size distribution is Pareto above the threshold of 9 Million Euros, as the logarithm of density is a straight line when firm size is Pareto distributed. The distance between the Pareto benchmark and our data can be interpreted as the “missing firms” in the sample. In order to account for these missing firms, we assume that the real distribution of firms g(ω) is the Pareto distribution estimated in Figure 2, and suppose that unobserved firms are fully informal in 2009 and remain fully informal after the tax increase. This assumption is a compromise between two extreme assumptions: 1. that we observe all firms, and 2. that the missing firms are similar (in terms of transparency and leverage) to the observed ones. As a robustness check, we compute our main quantities of interest in both cases, and use the results as reasonable bounds for the true elasticities. Another question that arises is whether we observe the actual endowment of firms or whether this variable already suffers from under-reporting. In the model, taxes are not directly based on firm endowment but on value added, and we suppose that firm endowment is fully observed by tax authorities. In order to be consistent with the model, we consider that the assets reported in Hellastat reflect total firm size including assets that could be related to undeclared activity. By contrast, one can think that reported assets are assets in declared plants in which case we would need to consider that the observed firm size distribution is an endogenous object that is (slightly) different from reported firm size distribution because of misreporting. We describe in the online Appendix how we calibrate the model, and match the most important empirical moment—the distribution of leverage across firms. Table 3 reports the benchmark calibration. We later show the sensitivity of our results to these parameters. At the initial equilibrium, the level of aggregate transparency in the economy, defined as the ratio between the aggregate tax base and aggregate output, is equal to 0.82. This is slightly higher than what is typically estimated in the literature,24 and we may underestimate the influence of small firms in our analysis. However, 23

There is non-negligible exit in the recession, mainly driven by small firms with a higher-thanaverage leverage. We can perform the same exercise on the balanced panel without significant differences in the results. 24 The shadow economy in Greece is typically estimated around 25%. See Schneider et al. (2010).

18

those informal firms typically do not respond to changes in tax conditions—they form an inelastic informal sector. Accounting for these firms boils down to adding a fixed informal sector, which would mechanically reduce our estimates for aggregate transparency. 4.2

Quantitative results

Using our benchmark calibration, we analyze the effect of changes in the VAT rate on our economy. The objective of our numerical simulations is to replicate the 2010 Greek fiscal consolidation and analyze how the transparency response could explain the observed misalignment between predicted tax revenues and actual tax revenues. To this purpose, we set the same tax rates as the government and estimate our predicted tax revenues, and the elasticities (εγ , εv ). We update the VAT rates according to the austerity measures implemented in 2010. The minimum VAT rate increased from 4.5 to 5.5%, the reduced VAT rate from 9 to 11% and the standard VAT rate from 19 to 23%. The repartition along VAT categories is invariant with firm size. In practice, we run three experiments for firms subject to the low, medium and high tax rates and we aggregate our results using as weights the shares of firms producing goods subject to each VAT regime in order to deduce the aggregate response of the economy. The results are reported in the second column of Table 4. Following the increase in the tax rates, the model predicts a drop in the tax base of 9% explained by a decrease of transparency (−7%) and output (−2%). Given the amplitude of both responses (essentially the transparency adjustment), almost half of the increase in taxes is diluted and does not translate in higher tax revenues. We can interpret these results in terms of elasticities to taxes. We find that the elasticity of tax revenues to the change in VAT rate introduced by the austerity plans is ετ γv = 0.56. The model-based behavioral response is composed of two elements, the standard behavioral response with a decrease in the real activity, and the decrease in the extent to which the activity is declared. We estimate the second element to be the largest: the transparency response εγ accounts for a bit more than three quarters of the fall in the tax base (−0.34 out of −0.44), whereas the output response εv accounts for the remaining quarter (−0.10 out of −0.44). The elasticity of transparency to tax rates εγ is consistent with (i) the systematic empirical analysis conducted in Section 2 which gave a range between −0.2 (high legal enforcement) and −0.5 (low legal enforcement), and (ii) the empirical evidence on the elasticity of tax compliance to the Greek tax reform in 2010 (see online Appendix).25 We 25

The comparison between the empirical elasticities and the model predictions is not straight-

19

interpret this finding as an external validation of our calibration strategy, since we did not use this moment (the observed fall in VAT compliance) to calibrate the model. Since we acknowledge that there may be “missing firms” in our sample, we assume that we do not observe informal small firms in 2009 which remain fully informal after the tax increase. We now modify this assumption and rather consider that in our sample we observe all firms. Under this assumption, the elasticity of transparency and output are very similar to our benchmark case (respectively −0.32 and −0.11). In contrast, when we assume that there exist unobserved small firms that behave exactly like the observed ones, the absolute elasticity of transparency increases significantly (εγ = −0.48), because there are more firms responding by adjusting their transparency. One can therefore think that the elasticity of transparency should lie between these two extremes −0.48 < εγ < −0.32. As regards the elasticity of output, it remains almost unchanged in both cases. We have shown that most of the drop in expected tax revenues come from a drop in transparency. This transparency adjustment has also an impact on the extent to which output decreases. Indeed, when small and medium-size firms reduce their transparency, they tighten even further their credit constraints and reduce accordingly their credit demand. A simple experiment which highlights the quantitative impact of such channel is to replicate the fiscal consolidation maintaining constant the transparency of firms. Under the assumption of fixed transparency, the contribution of transparency to output changes is nil, i.e., νγ = 0. It allows us to identify νv = εv , i.e., the standard fall in output purged of the transparency effect. The last column of Table 4 reports the results of the simulation where the transparency response is shut down, that is when εγ = 0 and νγ = 0, and the only effect that is captured is the standard fall in output νv = −0.05. This result shows that the indirect impact of transparency on the output response accounts for more than half of the total output response. In other words, had the transparency been insensitive to changes in taxes, the output drop would be 1 percentage point lower. This simple exercise points to the large influence of the transparency channel both in the relatively small increase in tax revenues and in the subsequent output drop. In addition to the aggregate estimates, it is interesting to study the distributional implications of the tax hike. Figure 5 shows the elasticities of transparency and outforward. Indeed, the theoretical elasticity can be interpreted as causal as we keep fixed other characteristics of the economy (financial development, endowments, pre-tax relative returns to technology) while the empirical elasticities may partly reflect omitted variables, reverse causality or anticipations from economic agents. In addition, the empirical exercise relies on an average response across different economies with different fundamentals and different tax reforms (including changes in tax base which we do not model theoretically).

20

put to tax hike along firm size. Most of the drop in tax revenues is due to middle-size firms that either drop off the formal economy or adjust their transparency downward. In order to understand why the response of middle-size firms is important, we can represent our economy as follows. There are three types of firms: small informal ones, large transparent ones and middle-size firms. Following the tax hike, small firms remain informal and large firms remain transparent. If there were only such firms in the economy, there would be no transparency response to tax increase but only an output response driven by lower expected returns, and the overall elasticity of tax revenues to tax revenues would be close to 1. By contrast, middle-size firms react by changing their level of transparency, i.e., either by becoming fully informal or by reducing the extent to which they declare their activity. Tax base decreases so much for these firms that the elasticity of tax receipts to tax rates is negative. These firms are on the right-hand side of the Laffer curve: if there were only such firms in the economy, the increase in taxes would actually reduce tax revenues. These observations rationalize the use of size-dependent tax policies, a possibility that we further explore in Section 5. We also find direct evidence of this size-specific credit crunch in our panel of firms (see Figure 1): there has been a shift of credit out of small and medium-size firms during the crisis. The empirical evolution of credit along firm size is comparable to our theoretical predictions (see Figure 3), both quantitatively and qualitatively.26 Overall, our model does not only match well the aggregate evolution in credit and transparency, but also its distributional features. 5

Counterfactual experiments

In this section, we perform some comparative statics and estimate how variations in the fundamentals of our economy may explain wide variations in elasticities across countries, as illustrated by our empirical analysis. We also explore the impact of other tax policies in order to further analyze the heterogeneous response along firm size. 5.1

The role of fundamentals

We now investigate which fundamentals of our economy may explain the large observed differences in the tax compliance response to tax hikes. The previous analysis 26

In the online Appendix, we provide some evidence on the relationship between firm size and tax compliance elasticities. We focus on the 2010 Greek austerity plan, estimate regional tax compliance elasticities and show the importance of firm size differences across regions to predict the response of tax compliance.

21

shows that the elasticity of tax revenues to tax reforms depends on the number of firms at the margin between informality and formality, i.e., the number of firms relying on external finance but not fully transparent. The number of such firms is determined by (a) the range in which firms are almost indifferent between informality and access to credit, (b) the density of firms in this range, and both quantities are pinned down by fundamentals of the economy, i.e., the legal enforcement (represented here by lender protection) and firm size distribution. In the following lines, we describe our comparative statics exercise to understand the quantitative role of these two fundamentals. In our framework, changes to fundamentals modify both the “steady state” of the economy and the elasticities of tax revenues, transparency and output to taxes. We take advantage of this observation in order to represent our elasticities, not as a function of each underlying parameter but rather as a function of steady-state quantities. For the share λ of observed investment that can be pledged, we define λ 7→ Γ(λ), where Γ(λ) is the aggregate transparency. We compute the elasticities of tax revenues γτ v (λ), transparency γ (λ) and output v (λ) to taxes as functions of λ. The objective is to study how the response to taxes depends on steady-state aggregate transparency when the variations in steady-state aggregate transparency are only driven by different credit market conditions. Similarly, we compute the same elasticities as functions of ψ, and define the same mapping ψ 7→ Γ(ψ) for the aggregate transparency as function of the shape of the firm size distribution ψ. Figures 6 and 7 display these elasticities. The solid blue line is the elasticity of tax receipts (ετ γv = 1 + εγ + εv ), the dashed red line is the transparency component of the elasticity of tax receipts (1 + εγ ). On the horizontal axis, we report aggregate transparency which moves with the underlying fundamental: it increases with the pledgeability parameter λ, whereas it decreases with the shape parameter ψ. A high level of ψ corresponds to lower tail for the firm size distribution, and the density of firms at the margin of informality is higher. As shown in figure 6, as we increase the extent to which collateral can be pledged (λ), the elasticity of tax revenues to tax rate slightly increases. When financial development increases, the pressure of the credit constraint is lower for larger firm but higher for smaller firms, which are now investing more in the modern technology. Both effects together imply a higher aggregate elasticity γτ v for more financially developed economy (medium-size firms represent a low share of the economy). The variation in transparency response across the different institutional environments is quantitatively relevant: an increase from 78% to 82% in aggregate transparency due to a change in credit constraints reduces the transparency elasticity from -.40 to

22

-.25 (a variation that is of the same order of magnitude as our observed empirical estimates). By contrast, the output response to taxes seems to be barely affected by an improvement in financial development as the gap between the blue line (εγτ v ) and the dotted red line (1 + εγ ) in figure 6 remains constant. We then study how the relative weight of large versus small firms in the economy modifies the elasticity of tax revenues to tax reforms. An economy with a fat-tail firm size distribution (low ψ) is less responsive to taxes since most of the effect comes from the weight of medium-size firms. In parallel, the output response increases, as the number of unconstrained and large firms increases (these large firms are the ones for which the standard behavioral response to taxes is the largest). This is the reason why the gap between the dotted red line (1 + εγ ) and the blue line (εγτ v ) in figure 7 widens and the overall elasticity of tax revenues to taxes slightly decreases with aggregate transparency. If we only increase instead the number of medium-size firms, we would find a much lower aggregate elasticity εγτ v . To summarize, while the elasticity of tax revenues to tax reforms is not sensitive to changes in firm size distribution, the transparency response is quite responsive: an increase from 82% to 86% in aggregate transparency due to a change in firm size distribution (lower number of small firms) reduces the transparency elasticity from -.35 to -.28. Finally, in figure 8, we plot the aggregate elasticity for different combinations of λ and ψ while keeping the sanctions at its benchmark level. The quantitative results suggest that the elasticity of tax revenues to tax reforms is in the range [0.45, 0.65]. These results are consistent with the empirical estimation of the changes in VAT compliance associated with VAT reforms shown in Table 2. The lower bound of the range of elasticity corresponds to an economy where financial development is low and the firm size distribution is shifted towards small and medium size firms. The upper bound refers instead to countries where legal enforcement is of better quality and firms on the verge of becoming informal are very small.27 This simple analysis points to the distribution of firm size as a crucial, and so far, under-explored, factor behind the success of a fiscal consolidation in raising tax revenues. 5.2

The characteristics of tax reforms

The previous section has provided comparative statics along some of the fundamentals governing the steady-state economy. The next exercise examines the sensitivity 27

Many developed countries have a firm size distribution which may be represented as a Pareto with a shape parameter close to 1, as shown by Axtell (2001) for the US or Levchenko et al. (2010) for France.

23

of the response to the type of tax reforms. Indeed, while most peripheral European countries in the recent crisis have increased VAT rates to a certain extent, some economies have relieved tax pressure on small and medium firms. For instance, the 2010-2011 VAT reform in Portugal (with an increase in the standard rate from 20 to 23% between 2010 and 2011) was followed by a size-dependent reduction in corporate tax rates, and a simplified corporate tax reporting system for smaller firms. In Spain, a similar phased-in reduction in corporate tax rates also featured reduced rates for small firms, and Italy raised the VAT registration threshold for smaller businesses. One interesting feature of these fiscal reforms is their size-dependence. The rationale behind our current model should justify such policies. Indeed, as shown in Figure 5 and briefly discussed in Section 4, the response to a VAT increase crucially depends on firm size, and size-dependent policies may be more desirable than an unconditional increase. A targeted VAT exemption for small firms, e.g., a stable rate, should neutralize both margins of the transparency response: the choice of transparency for entrepreneurs in the modern sector would not be further distorted, and the relative returns to the modern technology would be left unchanged. A reform of corporate tax rates as in Portugal or Spain or a simpler reporting system would mostly change the relative returns to the modern technology and alleviate the distortion at the extensive margin, but would not affect the choice of leverage in the modern sector. In order to quantify these effects, we proceed as follows. We build upon the calibrated baseline economy and the baseline VAT reform (see Section 4), and we perform the following sensitivity experiments. In a first step, we add to the baseline experiment the equivalent of a reduction in corporate tax rates as observed between 2013 and 2014 in Portugal, with a decrease from 25 to 23% for larger firms and a further reduction to 17% for firms with an endowment below 1 Million.28 Experiment (E1) thus features the same VAT increase as in the baseline, and a reduction in corporate taxes. Column 2 of table 5 reports the elasticities of VAT revenues, output and transparency to the VAT rate following experiment (E1). While the corporate tax reduction alleviates the transparency and output responses, its quantitative impact is negligible. The main explanation behind this pattern is that corporate taxes do affect the relative returns to the modern technology, but not the transparency choice and access to credit conditional 28

The threshold for reduced corporate rates in Portugal was based on profits before taxes. In order to convert a threshold based on profits into a threshold based on endowment, we use Hellastat and choose the maximum endowment for which the median profit is below the Portuguese threshold of 12’500 euros.

24

on using the modern technology. A reduction thus pushes more firms to invest in the modern technology but the transparency choice is unaffected. In a second step, we study two additional experiments to decompose the lower response of transparency and output observed in experiment (E1). Experiment (E2) features a uniform decrease in corporate rates from 25 to 23%, while experiment (E3) only features the reduction for small firms (from 25 to 17%). Columns 3 and 4 of Table 5 show that most of the attenuation effect observed in experiment (E1) was related to the global reduction in corporate rates (as opposed to the additional reduction for small firms). These findings confirm, as discussed earlier, that a nonnegligible source of distortion remains, even with low corporate taxes: entrepreneurs with low endowment still face high credit constraints and thus low expected returns to the modern technology. In a third step, we allow the government to implement a size-dependent VAT reform.29 We assume that the government increases VAT as in the baseline but only above a certain endowment threshold that we set to 1 Million Euros, and smaller businesses face the pre-reform VAT rates. As shown by experiment (E4) in Table 5, the transparency response to the VAT increase is then reduced by a third, and the output response is 15% smaller than in the baseline. We then vary the threshold between 0 and 40 million euros, and replicate experiment (E4) with the varying thresholds. The findings are shown in Figure 9. The left panel reports the elasticity of tax revenues to the VAT increase as a function of the threshold, and the right panel the percentage change in tax revenues. As the elasticity of tax receipts is negative for very small firms (see Figure 5), the aggregate elasticity of tax revenues to tax rates initially increases as the policy exempts small firms from the tax hike. It reaches its maximum when all firms for which the elasticity is negative are exempted. This finding shows that size-dependent VAT reforms are more effective than a uniform VAT hike, and there is an “efficient” threshold: a revenue-maximizing policy should exclude small and medium firms with total assets between 0 and 15 million euros. These findings shed some light on reforms in other peripheral European economies. Size-dependent tax reforms echo the heterogeneous response of tax revenues to tax rates along firm size. The size-dependent corporate tax reforms in Portugal or Spain marginally relieved tax pressure on these very responsive firms. However, and as shown by the last counterfactual exercise, the need for a size-dependent policy is particularly pronounced through VAT because VAT distorts demand for credit in addition to tilting the trade-off between the traditional and modern sectors. 29

This type of tax reforms usually hinges on differential changes in standard, reduced and superreduced rates, or a registration threshold.

25

6

Conclusion

When firms adjust the degree to which they declare their activity—their transparency— in response to tax changes, the standard erosion of the tax base is augmented by the erosion of transparency. We document this effect using a novel measure of VAT compliance across 35 countries. We then develop a model in which firms adjust their declared activity. We calibrate the model to the recent fiscal consolidation episode in Greece and show that there is a large behavioral response and three quarters of the overall behavioral response to the tax increase come from this transparency channel. As transparency guarantees a better access to credit market, firms switching to the informal sector are excluded from credit markets thereby depressing aggregate output. One direct implication is that the amplitude of the transparency response depends upon the number of firms at the margin between formality and informality. The behavior of those firms is very sensitive to changes in the trade-off between credit access and tax compliance. Low tax monitoring and intermediate financial development contribute to having quite large and numerous small-medium firms for which the transparency response to taxes is important. Our quantitative analysis shows that variations in the fundamentals of the economy, i.e., the legal enforcement, the financial development and the firm size distribution, are able to explain the large differences across countries in the response of tax compliance to tax hikes shown in our empirical results. Quantitatively, we find leakages consistent with our empirical estimates. In our quantitative framework, the increase in tax revenues is almost twice lower than if tax compliance had been constant. The policy implications of our analysis are not immediate. We show that a drastic fiscal consolidation in an economy with low tax enforcement and low financial development is very likely to be diluted. Improving these institutions would help but is a difficult task: periods of economic turbulence may not be times in which structural reforms are simple to implement. One direct implication of the model is that the efficiency of a tax increase essentially depends on the number of firms (and their size) that are almost indifferent between declaring their activity or being fully informal. Economies with low financial development and inefficient tax monitoring may be willing to implement size-dependent tax policy and exempt very responsive firms. One possible policy would consist in designing targeted tax deductions when firms rely on external finance, or simply conditioning tax payments on firm size. Finally, there are several macroeconomic mechanisms that we ignore in our quantitative exercise. For instance, the analysis relies on a partial equilibrium model,

26

and the supply of capital, the unique production factor, is infinitely elastic. Recent episodes of fiscal consolidation, e.g., Greece in 2010, were a response to debt overhang. As the domestic banking sector usually owns a large share of sovereign bonds, fiscal policies may indirectly affect credit supply through this default risk channel. The analysis is also in closed economy, and assumes that entrepreneurs are in fixed supply. An open economy framework could account for the sensitivity of foreign direct investment to taxation, which may not be negligible as shown in Cummins and Hubbard (1995). Last, our model ignores the motivation behind fiscal consolidations. An interesting extension would consist in endogenizing government decisions. Reducing debt levels, a positive aspect of fiscal consolidations, may come at the expense of economic viability, especially for smaller firms. A government with low financial development and loose tax enforcement would have lower incentives to implement fiscal consolidations, and higher incentives to default. We leave these extensions for further research. References Alesina, Alberto F. and Silvia Ardagna (2009), Large changes in fiscal policy: Taxes versus spending, Working Paper 15438, National Bureau of Economic Research. Alesina, Alberto, Omar Barbiero, Carlo Favero, Francesco Giavazzi and Matteo Paradisi (2015), ‘Austerity in 2009-13’, Economic Policy 30(83), 383–437. Andreoni, James, Brian Erard and Jonathan Feinstein (1998), ‘Tax Compliance’, Journal of Economic Literature 36(2), 818–860. Artavanis, Nikolaos, Adair Morse and Margarita Tsoutsoura (2016), ‘Measuring Income Tax Evasion Using Bank Credit: Evidence from Greece’, The Quarterly Journal of Economics 131(2), 739–798. Auerbach, Alan J. and Yuriy Gorodnichenko (2012), ‘Measuring the Output Responses to Fiscal Policy’, American Economic Journal: Economic Policy 4(2), 1– 27. Axtell, Robert L. (2001), ‘Plants and productivity in international trade’, Science 293(5536), 1818–1820. Cai, Hongbin and Qiao Liu (2009), ‘Competition and Corporate Tax Avoidance: Evidence from Chinese Industrial Firms’, Economic Journal 119(537), 764–795.

27

Capasso, Salvatore and Tullio Jappelli (2013), ‘Financial development and the underground economy’, Journal of Development Economics 101, 167–178. CPB (2013), Study to quantify and analyse the VAT Gap in the EU-27 Member States, Final report, CPB Netherlands Bureau for Economic Policy Analysis, European Commission. Cummins, Jason and R. Glenn Hubbard (1995), The Tax Sensitivity of Foreign Direct Investment: Evidence from Firm-Level Panel Data, in ‘The Effects of Taxation on Multinational Corporations’, NBER Chapters, National Bureau of Economic Research, Inc, pp. 123–152. Desai, Mihir A., Alexander Dyck and Luigi Zingales (2007), ‘Theft and taxes’, Journal of Financial Economics 84(3), 591–623. Djankov, Simeon, Oliver Hart, Caralee McLiesh and Andrei Shleifer (2008), ‘Debt enforcement around the world’, Journal of Political Economy 116(6). Ellul, Andrew, Tullio Jappelli, Marco Pagano and Fausto Panunzi (2015), ‘Transparency, tax pressure, and access to finance’, Review of Finance 20(1), 37–76. Enste, Dominik H. and Friedrich Schneider (2000), ‘Shadow Economies: Size, Causes, and Consequences’, Journal of Economic Literature 38(1), 77–114. Favero, Carlo, Francesco Giavazzi and Jacopo Perego (2011), ‘Country Heterogeneity and the International Evidence on the Effects of Fiscal Policy’, IMF Economic Review, Palgrave Macmillan 59(4), 652–682. Fisman, Raymond and Shang-Jin Wei (2004), ‘Tax Rates and Tax Evasion: Evidence from Missing Imports in China’, Journal of Political Economy 112(2), 471–500. Fortin, Bernard, Nicolas Marceau and Luc Savard (1997), ‘Taxation, wage controls and the informal sector’, Journal of Public Economics 66(2), 293–312. Garicano, Luis, Claire Lelarge and John Van Reenen (2016), ‘Firm size distortions and the productivity distribution: Evidence from france’, The American Economic Review 106(11), 3439–3479. Gordon, Roger and Wei Li (2009), ‘Tax structures in developing countries: Many puzzles and a possible explanation’, Journal of Public Economics 93(7-8), 855– 866.

28

Ilzetzki, Ethan, Enrique G. Mendoza and Carlos A. V´egh (2013), ‘How big (small?) are fiscal multipliers?’, Journal of Monetary Economics 60(2), 239–254. Kleven, Henrik Jacobsen, Martin B. Knudsen, Claus Thustrup Kreiner, Soren Pedersen and Emmanuel Saez (2011), ‘Unwilling or Unable to Cheat? Evidence From a Tax Audit Experiment in Denmark’, Econometrica 79(3). Kumar, Krishna B, Raghuram G Rajan and Luigi Zingales (1999), What determines firm size?, Technical report, National bureau of economic research. Levchenko, Andrei A., Julian di Giovanni and Romain Ranci`ere (2010), Power laws in firm size and openness to trade: Measurement and implications, IMF Working Papers 10/109, International Monetary Fund. Pappa, Evi, Rana Sajedi and Eugenia Vella (2015), ‘Fiscal consolidation with tax evasion and corruption’, Journal of International Economics 96(S1), 56–75. Pomeranz, Dina (2015), ‘No Taxation without Information: Deterrence and SelfEnforcement in the Value Added Tax’, American Economic Review 105(8), 2539– 2569. Porta, Rafael La and Andrei Shleifer (2008), ‘The Unofficial Economy and Economic Development’, Brookings Papers on Economic Activity 39(2 (Fall)), 275–363. Rauch, James E. (1991), ‘Modelling the informal sector formally’, Journal of Development Economics 35(1), 33–47. Romer, Christina D. and David H. Romer (2010), ‘The macroeconomic effects of tax changes: Estimates based on a new measure of fiscal shocks’, American Economic Review 100(3), 763–801. Schneider, Friedrich, Andreas Buehn and Claudio Montenegro (2010), ‘New Estimates for the Shadow Economies all over the World’, International Economic Journal 24(4), 443–461. Slemrod, Joel and Shlomo Yitzhaki (2002), Tax avoidance, evasion, and administration, in A. J.Auerbach and M.Feldstein, eds, ‘Handbook of Public Economics’, Vol. 3 of Handbook of Public Economics, Elsevier, chapter 22, pp. 1423–1470. Straub, St´ephane (2005), ‘Informal sector: The credit market channel’, Journal of Development Economics 78(2), 299–321.

29

Tables Table 1. Elasticity of tax compliance to tax rates.

Panel A: Standard VAT rate VAT Compliance Elasticity β

Panel B: Effective VAT rate VAT Compliance Elasticity β

Country fixed effects Year fixed effects Economic cycle Government expenditures Sectoral composition Trade

(1) -.403∗∗∗ (.053) [605]

(2) -.355∗∗∗ (.072) [468]

(3) -.407∗∗∗ (.078) [421]

(4) -.403∗∗∗ (.078) [421]

(1) -.465∗∗∗ (.045) [605]

(2) -.403∗∗∗ (.061) [468]

(3) -.457∗∗∗ (.066) [421]

(4) -.455∗∗∗ (.066) [421]

Yes Yes Yes No No No

Yes Yes Yes Yes No No

Yes Yes Yes Yes Yes No

Yes Yes Yes Yes Yes Yes

Significantly different than zero at ∗ 90% confidence, ∗∗ 95% confidence, ∗∗∗ 99% confidence. Robust standard errors are reported between parentheses. The number of observations for each specification is reported between brackets. Each cell displays the estimate of a separate regression (see equation 1).

30

Table 2. Elasticity of tax compliance to tax rates – high versus low legal enforcement.

Panel A: Standard VAT rate VAT Compliance Elasticity βl Difference βh − βl

Panel B: Effective VAT rate VAT Compliance Elasticity βl Difference βh − βl

(1) -.412∗∗∗ (.076) .018 (.105) [605]

(2) -.468∗∗∗ (.092) .282∗∗ (.143) [468]

(3) -.585∗∗∗ (.097) .470∗∗∗ (.157) [421]

(4) -.583∗∗∗ (.097) .479∗∗∗ (.156) [421]

(1) -.458∗∗∗ (.060) -.014 (.085) [605]

(2) -.461∗∗∗ (.073) .185 (.127) [468]

(3) -.556∗∗∗ (.078) .324∗∗ (.139) [421]

(4) -.556∗∗∗ (.079) .331∗∗∗ (.139) [421]

Yes Yes Yes No No No

Yes Yes Yes Yes No No

Yes Yes Yes Yes Yes No

Yes Yes Yes Yes Yes Yes

Country fixed effects Year fixed effects Economic cycle Government expenditures Sectoral composition Trade

Significantly different than zero at ∗ 90% confidence, ∗∗ 95% confidence, ∗∗∗ 99% confidence. Robust standard errors are reported between parentheses. The number of observations for each specification is reported between brackets. Each column in each panel displays the estimate of a separate regression (equation 1 augmented by the interaction with a dummy high-enforcement country).

α

Table 3. Austerity in Greece – benchmark calibration. Interpretation Value Rationale Returns to scale 0.82 Sales - Hellastat (2009)

r

Interest rate

0.08

Bank of Greece (2009)

A

Productivity factor

0.92

Distribution output - Hellastat (2009)

λ βp c0

Collateral pledgeability Innovation (scale) Innovation (factor)

0.50 0.30 2.10

Distribution leverage - Hellastat (2009) Distribution leverage - Hellastat (2009) Distribution leverage - Hellastat (2009)

ψ

Shape (size dist.)

1.9

Distribution size - Hellastat (2009)

θ

Punishment

1.5

Tax Procedure Code (2010)

τ

VAT - low rate VAT - medium rate VAT - high rate

.045 (18%) .09 (12%) .019 (70%)

VAT - Greece (2009) VAT - Greece (2009) VAT - Greece (2009)

Note: see Section 4 and the Appendix Section B.3 for a detailed description of the calibration.

31

Table 4. Austerity in Greece – results.

Tax rate Tax base Output Transparency Tax revenue ετ γv Transparency εγ Output εv

Baseline experiment Fixed Percentage changes 21.410 -9.218 -2.073 -7.341 Elasticities 0.557 -0.343 -0.099

transparency 21.410 -1.390 -1.065 -0.332 0.951 0.000 -0.049

Note: see Section 4 for description of the experiments.

Table 5. Austerity in Greece – counterfactual reforms.

Tax revenue ετ γv Transparency εγ Output εv

Baseline

E1

0.5573 -0.3435 -0.0992

0.5653 -0.3429 -0.0918

E2 Elasticities 0.5632 -0.3431 -0.0937

E3

E4

0.5601 -0.3433 -0.0967

0.6793 -0.2359 -0.0848

Note: Experiment E0 represents the baseline experiment, i.e., an increase of 21.4% in the effective VAT rate. Experiments E1-E3 add to this experiment a reduction in corporate tax rates: 25 to 23% for large firms and 25 to 17% for small firms in E1, 25 to 23% for all firms in E2, 25 to 17% for small firms in E3. E4 builds on the baseline experiment (no change in corporate tax rates) and exempts small firms from the VAT increase. Small firms are firms with an endowment lower than 1 Million euros.

32

Figures Figure 1. Leverage as a function of firm size before and after the 2010 tax reform.

Note: Source Hellastat, 2009, 2011. This graph displays the leverage by firm size (total assets) before (2009) and after (2011) the austerity plan.

Figure 2. Firm size distribution Greece.

Note: This figure represents the observed firm size distribution in Hellastat (2009) and the predicted density computed using only firms with endowment above 10M euros. The x-axis is on a logarithmic scale.

33

Figure 3. Leverage and transparency : the impact of the 2010 tax reform.

(a) Leverage.

(b) Transparency.

Note: Leverage and transparency along firm size for the benchmark calibration (solid line) and the 2010 austerity plan simulation (dashed line).

Figure 4. Empirical and theoretical leverage and output.

(a) Leverage.

(b) Output.

Note: Benchmark calibration. The solid black lines are the calibrated leverage and output, the dashed blue lines are the empirical leverage and output for firms with assets between 0.5 and 50M euro (smoothed using a HP filter).

34

Figure 5. Transparency and output elasticity by firm size.

Note: The solid line is the elasticity of transparency εγ , the dashed line is the elasticity of output εv as a function of firm size. Both are computed using the 2010 austerity plan simulation. See Section 4 for a description of the experiment.

Figure 6. Elasticity of tax revenues to tax reforms : the role of credit frictions.

Note: The solid blue line is the elasticity of tax revenues (ετ γv = 1 + εγ + εv ), the dashed red line is the transparency component of the elasticity of tax revenues (1+εγ ). In the horizontal axis we report the aggregate transparency Γ(λ) which is associated with values of λ ∈ [0.42, 0.58]. See Section 5 for a description of the counterfactual experiment.

35

Figure 7. Elasticity of tax revenues to tax reforms : the role of firm size distribution.

Note: The solid blue line is the elasticity of tax revenues (ετ γv = 1 + εγ + εv ), the dashed red line is the transparency component of the elasticity of tax revenues (1+εγ ). In the horizontal axis we report the aggregate transparency Γ(ψ) which is associated with values of ψ ∈ [1.3, 2.3]. See Section 5 for a description of the counterfactual experiment. Note that aggregate transparency is decreasing with the shape of firm size distribution ψ.

Figure 8. Aggregate elasticity of tax revenues – credit frictions λ and firm size shape ψ.

Note: This Figure represents the aggregate elasticity of tax revenues as a function of credit frictions λ and firm size shape ψ. See Section 5 for a description of the counterfactual experiment.

36

Figure 9. Aggregate elasticity of VAT revenues and percentage change in VAT revenues – VAT rise exemption below a threshold.

(a) Elasticity of tax revenues.

(b) Change in tax revenues.

Note: Response to counterfactual VAT reforms for which smaller firms are exempt from the VAT increase. For instance, for a threshold of 1 Million euros, all firms below 1 Million euros would be subject to the pre-reform VAT rates. See Section 5 for a description of the counterfactual VAT reforms.

37

Online Appendix A

Measuring tax compliance

In this section, we describe a series of challenges and cleaning procedures that we implement to get closer to the ideal measure for the largest set of countries/years. A.1

Time aggregation

First, we are interested in the short-term fluctuations of the measure T Ct,c , and we need to smooth the “high-frequency” measurement error. Tax reforms are often implemented during the year, while national accounts are closed at the end of each period, i.e., year or quarter. For this reason, we collect the effective tax rate for a unit of consumption in any given month. When tax rates were changed during the course of the year, we use these monthly measures and construct the annual effective tax rate by weighting each tax rate by the consumption observed during its spell. When consumption could not be observed at a higher frequency than the period, we construct the annual effective tax rate by weighting each tax rate by the duration within the period during which it was enforced. Given that we rely on monthly VAT data, about 1/12 of total consumption could be allocated to the wrong VAT rate in the worse case scenario. This time aggregation issue thus generates little residual noise in our estimates. A.2

Measurement error in tax base

Second, some tax reforms do not modify rates but also modify the category of goods that are subject to the different tax regimes. For instance, for countries entering in the European Union, art galleries would pass from category 1 to category 3. In such instances, we redefine our tax base correctly when our decomposition in the different categories allows us to observe exactly the category that has been modified. When, instead, we do not observe consumption in art galleries but we observe consumption for a larger category (“cultural goods”), we reconstruct a synthetic tax base for art galleries and the other cultural goods by considering the share of art galleries among cultural goods in specific years or in some benchmark country when more detailed consumption categories are never documented. Along the same lines, VAT can be collected for all registered firms or there may exist a minimum threshold. In the case of a reform, we would recreate the new tax base by subtracting the average share of value added created by firms below the threshold. Remark that it is likely that the

38

actual share of value added reacts to the changes in tax coverage, a response that we mostly ignore. These adjustments for changes in categories and exemptions may remain subject to measurement errors. With our correction on tax categories, we may still attribute a share of a certain expenditure category to the wrong tax rate. This share would be the residual consumption of a specific good compared to its benchmark consumption as computed either in a specific year, or for the United States, and may not be negligible. However, within this measurement error, only a small fraction should correlated with changes in the tax reforms: in our example, we would attribute to art galleries their share under another tax regime than the one induced by the tax reform and actual consumption may be lower than imputed consumption due to the behavioral response. To give orders of magnitude, if the consumption of an exempted good that represents half of a category increases by 10% following a tax reform while the imputed consumption is fixed, we would misclassify 5% of the expenditure to exempted VAT at the category level. A.3

Accounting for structural reforms

Third, some reforms modify the tax environment without modifying the tax rates per se. For instance, online registration considerably simplifies the registration process. We collect this information and clean for the jumps associated with such reforms. This correction implies that our measure is a relevant measure to study the cyclical behavior of VAT compliance, but it may not account for some structural reforms which affect the long-term levels.

39

Austerity in Greece

B B.1

Empirical evidence on the response of tax compliance to tax reforms

In this section, we study the average response of tax compliance following the Greek austerity plan of 2010 and we examine the heterogeneity of such response across the intensity of the tax shock.30 The 2010 tax reform in Greece essentially consisted of a VAT increase from a standard rate of 19% to 21% and finally 23% after a revision in July 2010, and an increase in the excise on unleaded petrol from 36 cents per liter to 61 and finally 67 cents after a similar revision. For these two taxes, VAT and excise on unleaded petrol, we plot the measure of tax compliance over the period 2007-2012 in Figure B1. One important fact stands out. In 2010, while the VAT and excise rates markedly increase in Greece (by respectively 20% and 85%), VAT and excise compliance drop by approximately 6.4% and 10%, which corresponds to elasticities of VAT and excise compliance of −.32 and −.12. The fact that the tax base shrinks by 6% for VAT whereas it remains almost unchanged for the excise points to a different behavior of tax compliance in response to these tax hikes. The payment of excises is indeed generally better enforced, and it is much more difficult to conceal or under-report than the VAT payment. We now show that there exist large variations in this response across regions, and these variations can be related to the region-specific tax shock. We use Elstat data to observe (i) the annual VAT revenues (total and subtotals for legal entities and individuals), and (ii) the annual value added in each 1-digit industry between 2006 and 2011 for 51 regional units. We can then proceed as described in the previous subsection, and compute the regional VAT compliance T Cj,t , for each region j. There is a difference with the previous exercise that is worth noting: we now observe consumption for industries and not goods, and only at the 1-digit level. Accordingly, our matching between the observed units for consumption and the defined categories for VAT rates is imperfect and relies on some assumptions: we construct the incidence of each 2-digit good category in the different 1-digit industry category at the national level, and we associate to each 1-digit industry an “effective” VAT rate at the regional level. Doing so, we ignore all intra-industry reallocation across goods subject to different tax rates, which may bias our measure of tax compliance. However, we can bound the bias generated by such composition effect: within an industry category taxed on average at 16%, a substitution of 5 percentage points between good A taxed at 13% and good B taxed at 23% would 30

For a detailed description of the VAT hikes see Table B1.

40

generate a fluctuation of 0.05×10/16 ≈ 0.03 in the tax compliance measure, which is an order of magnitude lower than the standard deviation of fluctuations in regional VAT compliance.31 In order to illustrate the differential response across regions, we compute ∆T Cj , i.e., the gap (in percentage points) of the regional VAT compliance between the pre-reform period (2006-2009) and the post-reform period (2010-2011). We observe a large cross-regional variation in the evolution of VAT compliance before and after the austerity plan. The standard deviation of ∆T Cj across regions is around 0.12 implying a difference of 0.15 points between the first and last quartile of regions. Some regions experience a marked decrease in their compliance, namely Thessaloniki and Attiki, which are home to the two largest cities, whereas some others experience a large increase, e.g., the islands Chios, Kyklades, Samos or Zakynthos.32 The large fall in compliance in Attiki and Thessaloniki, the two regions where about 65% of Greek GDP is generated, explains the aggregate fall in the economy. Below we provide two robust pieces of evidence that are common to the overall Greek economy and help understand (i) these regional disparities and (ii) which factors drive fluctuations in tax compliance. First, and similarly to our cross-country analysis (see table 1), the fluctuations in compliance are strongly related to the fluctuations in effective VAT rate, independently of fluctuations in output. We define the effective VAT rate as the average tax rate on a unit of output produced in the region. This tax rate would be 19% for a region whose industry is entirely dedicated to the production of category 1 goods. There exist large variations in the fluctuations in effective VAT rates due to the regional sectoral composition but also to the tax exemptions implemented in some regions (mainly islands). In the left panel of figure B2, we plot ∆T Cj as a function of fluctuations in effective VAT rate controlling for fluctuations in output. The regions where the VAT hike is larger are those with the larger drop in compliance on average. The relationship is extremely strong and robust to the addition of sector-specific output growth. Second, the other main predictor for the fluctuations in compliance before and after the tax hike is the average regional firm size (measured by total assets). The regions where the average firm size (measured by total assets) is larger are also those with larger drop in collection efficiencies on average, as shown in the right 31

At the national level, an increased incidence of 5 percentage points for a 2-digit good in the 1-digit industry is about twice as large as the standard deviation in the incidence. 32 There exists a long-standing lower tax regime that applies on the Aegean islands in order to foster tourism. Besides, tax enforcement is notably lower. In 2010, the Greek authorities decided to increase enforcement without revising the exemptions and we see that VAT revenues strongly responded in these islands.

41

panel of figure B2.33 These results are obtained cleaning for the variations induced by regional GDP growth and the regional growth rate in effective VAT. We summarize these two results in Table B2. In this table, we first show the unconditional correlation between firm size and changes in VAT compliance weighted by the regional value added (first column). The correlation is very large: a region with average firm size of 0.2 M euros experience a 10 percentage point decrease in tax compliance relatively to a region with average firm size of 0.1 M euros. We then add the change in the effective VAT rates. Both the firm size and the change in VAT rates are strong predictors for fluctuations in VAT compliance (they explain almost 70% of the total variation in this measure ∆T Cj ). We also condition this correlation by other important regional characteristics. We include the regional growth rate (third column) and the sector-specific growth rates (fourth column). All set of additional controls capture only a small part of total variations in ∆T Cj .34 Firm structure and variations in effective VAT rates are the only relevant regional characteristics which predict leakages following the 2010 reform. B.2

Tax compliance and credit access

Having established the link between tax hikes and subsequent tax compliance, we now explore the impact of such tax compliance on credit access. As long as hidden activity is not as pledgeable as declared activity, the response of tax compliance to the VAT hikes should imply a credit flow out of medium-size firms in 2010-2011. In turn, the stronger response of tax compliance in those regions where the share of medium-size firms is larger should also be associated to a stronger credit crunch. We investigate the impact of the response to the austerity plan on credit access using a panel of firm-level balance sheet data on a quasi-exhaustive sample of Greek firms35 . The panel dimension of our data allows us to follow the credit history of these firms and, in particular, assess the degree to which they rely on external finance. First, the fiscal adjustment in 2010 is associated with an overall decrease in leverage in Greece : for all given firm size, there is at least a small decrease in their access to credit. In Figure 1, we plot the leverage – the ratio of external funds to total assets – in 2009 (blue line) and 2011 (red line) by firm size for the whole country. 33

We exclude the Attic region from this picture because it has a much larger average firm size and tax compliance than the other regions. Including the Attic region would make the relationship even stronger. Source : Annual business registers, Elstat statistical yearbook 2010. 34 Our results are also robust to the addition of a dummy for Aegean islands, the addition of the Attic region and they are robust when we do not use any weights. We also control for regional sector-specific incidence captured by the employment shares, and GDP per capita. 35 Hellastat 2001-2013, see Section 4 for a detailed description of the dataset.

42

We keep in this figure all firms, including new entrants and exiters. Notice, however, that there is more exit during the recession (the exit rate in our dataset is 5% in 2009, 10% in 2010 and 15% in 2011) and the exiters have generally a slightly higher leverage. We thus later use the panel dimension to capture the leverage response cleaned of compositional effects. Second, Figure 1 shows the heterogeneous response of credit across firm size. The leverage is close to 0 in 2009 for firms with less than 2 million euros in total assets, and it remains negligible in 2011. Large firms with assets above 20 million euros have a leverage of about 0.33 in 2009 and it slightly decreases in 2011.36 In contrast, medium-size firms experience a substantial fall in their access to external finance. For instance, a firm with 10 million euros in total assets exhibit a leverage of 0.3 in 2009 and 0.25 in 2011. These results are robust to a large set of controls: e.g. even within the same industry, medium-size firms are the only ones whose leverage drops. The shift in leverage is slighty less pronounced in the balanced panel but the results are qualitatively similar. We report in Table B3 the panel estimation of the drop in leverage between 2007-2009 and 2010-2012, and we distinguish three bins of firms: small firms with asset below 2M euros over the period, small-medium firms with assets between 2 and 20 M euros and medium firms with assets between 20 and 50 M euros. The U-shape of the drop in leverage with firm size is confirmed in this panel specification. Small firms see their leverage decrease by 0.02, small-medium by about 0.04 and medium firms do not experience any decrease in leverage. Third, we show that the decrease in leverage is larger in regions where tax compliance is low. In Figure B4, we show the evolution of leverage in 2011 relatively to 2009 in regions with high versus low response of tax compliance. The downward shift in leverage is substantially larger in regions where the response of tax compliance to tax hikes is larger. We further explore this correlation in Figure B5. We plot the regional change in tax compliance ∆T Cj against its counterpart change in leverage for medium-size firms (with assets between 2 Million euros and 50 Millions euros) around the austerity plan. The correlation is positive and significant (the elasticity is 0.83 with a standard error of 0.28). Interestingly, the elasticity is close to 1, thereby supporting the idea that credit is proportional to declared activity for credit-constrained firms. While there may be some differential compositional effects across regions, they do not have an impact on this correlation. In the next section, on the basis of the three stylized facts presented above, we argue that the aggregate response of tax compliance to tax hikes is relevant in a 36

We do not expect very large firms to respond because they are subject to a much tighter monitoring from tax authorities and investors. We do not expect the leverage of very small firms to decrease because it is already very close to zero.

43

country with weak tax enforcement, because medium-size firms substantially reduce their declared activity. In turn, these firms lose part of their access to external finance, which reduces aggregate investment. For our last effect to exist, we need access to credit to be affected by the degree to which a firm activity is concealed to tax authorities. B.3

Calibration of the model to the austerity in Greece

We use our balance sheet data to calibrate the model. We start by estimating the parameters that are directly observed. We estimate the elasticity of sales with respect to their size for firms with sales above 0.1M Euros using a specification which controls for firm-specific characteristics. It is well-known that such estimations suffer from endogeneity bias that we cannot fully alleviate. However, both cross-firms and within-firm across-time estimates give similar results – respectively 0.8 and 0.82 (see the fit of the relationship in Figure B3). We set α equal to 0.82. In the same vein, we estimate the Pareto parameter ψ which matches the asymptotic distribution of endowments in our sample, and find that ψ = 1.9. Then, we use our dataset to measure the average tax pressure on firms. We use the sector classification used in the analysis of the profitability of firms to measure the average VAT rate in the economy. In our dataset, about 69.4% of firms produce goods in the high VAT regime (19%), whereas 12.4% of firms are subject to the middle VAT regime (9%) and the remaining 18.2% of firms is either subject to the low regime or exempted (4.5%).37 We then compute the aggregate elasticity of tax receipts in the economy as the weighted sum of the elasticities for each tax regime. The interest rate is set to r = 0.08 such as to match the average short-term interest rate to non-financial corporations as of May 2010. For the parameters of our model that relate to the credit market frictions and the productivity of firms, we use the firms’ balance sheet information provided by our dataset, and choose our underlying parameters such as to match the resulting leverage and the total output of firms. The parameters which determine the distribution of leverage are the collateral pledgeability λ, and the probability to require such access, which is tied with the probability to operate with the modern technology p(c) = ( cc0 )βp .38 Intuitively, λ determines the leverage for large firms which operate only with the modern technology. c0 and βp help characterize the slope 37

In our database, over the period, we observe 60’662 firm×year observations under the low VAT regime, 41’238 firm×year observations under the middle VAT regime and 231’114 firm×year observations under the high VAT regime. 38 We do not observe the investment in R&D, and we cannot calibrate our innovation costs parameters such as to match real investment.

44

and curvature for the leverage of small and medium-size firms as a function of firm size. The best way to understand the role of each parameter is to look at Figure 4 : the level of the plateau is essentially pinned down by the collateral pledgeability parameter λ, whereas the slope and concavity of the first part of the curve are determined by c0 and βp . We therefore set these parameters such to minimize the distance between the theoretical and the empirical leverage shown in the left panel of figure 4. Similarly, we set the productivity factor A such as that our theoretical output reproduces closely the empirical output as shown in the right panel of figure 4. Concerning the monitoring intensity, it is hard to collect evidence on the strategy of Greek tax authorities. The statistics on the monitoring activity by Greek tax authorities are available since January 2011, and as such they do not allow to observe potential changes in the strategy around the implementation of the tax reforms.39 On the one hand, the endogenous auditing described in section 3.3 predicts an increase in the effort of tax authorities at detecting undeclared activity, and a strenghtening of tax enforcement has also been part of the reforms asked by the Troika, as suggested by the data availability starting in 2011.40 On the other hand, the tax authorities may suffer a significant reduction in the resources available for their auditing activity during a recession. In the end, we therefore choose to calibrate the model with an exogenous monitoring intensity, which is a linear function of the firm endowment. With respect to the sanctions, we parametrize them as to match the minimum administrative sanctions for VAT evaders in Greece.41 We therefore set θ = 1.5. In our numerical exercise, we do not aim at matching the overall receipts from auditing because we do not observe them in Greece. However, both in the data and in our model, sanctions are quite low. They only act as a threat and whether we capture them well or not would be visible on our levels of transparency rather than on the actual receipts due to tax monitoring.

39

See http://www.gsis.gr/gsis/info/gsis site/PublicIssue/Statistics.html. A tighter monitoring can also be observed in Italy during the same period with a marked increase in tax controls. 41 See the Tax Procedure Code. Legal penalties are huge but in practice rarely implemented. 40

45

Tables Appendix Table B1. VAT reforms in Greece (2010-2011). Goods and services affected by VAT Reform period

Jan. 2010

Mar. 2010

Tax regime Jul. 2010

Jan. 2011

Sep. 2011

Subject to Standard rate in 2010

19

21

23

23

23

CP020: CP030: CP040: CP050: CP060: CP070: CP080: CP090: CP120:

19 19 19 19 19 19 19 19 19

21 21 21 21 21 21 21 21 21

23 23 23 23 23 23 23 23 23

23 23 23 23 23 23 23 23 23

23 23 23 23 23 23 23 23 23

9

10

11

13

13

9 9

10 10

11 11

13 13

13 23

9 9

10 10

11 11

13 13

13 13

9 9 9 9 9

10 10 10 10 10

11 11 11 11 11

13 13 13 13 13

6,5 23 23 13 13

9 9

10 10

11 11

13 13

23 6,5

Subject to Super-reduced rate in 2010

4,5

5

5,5

6,5

6,5

CP090: Recreation and culture CP094: Recreational and cultural services (50%) CP095: Newspapers, books and stationery

4,5 4,5

5 5

5,5 5,5

6,5 6,5

6,5 6,5

excl.

excl.

excl.

excl.

13

excl. excl.

excl. excl.

excl. excl.

excl. excl.

13 13

excl. excl. excl. excl.

excl. excl. excl. excl.

excl. excl. excl. excl.

excl. excl. excl. excl.

excl. excl. excl. excl.

excl. excl.

excl. excl.

excl. excl.

excl. excl.

excl. excl.

Alcoholic beverages, tobacco and narcotics Clothing and footwear Housing, water, electricity, gas and other fuels Furnishings, households equipment Health Transport Communications Recreation and culture Miscellaneous goods and services

Subject to Reduced rate in 2010 CP010: CP011: CP012: CP040: CP044: CP045: CP060: CP061: CP062: CP063: CP073: CP094: CP110: CP111: CP112:

Food and non-alcoholic beverages Food Non-alcoholic beverages Housing, water, electricity, gas and other fuels Water supply and miscellaneous (50%) Electricity, gas and other fuels Health Medical products, appliances and equipment (50%) Out-patient services (50%) Hospital services (50%) Transport services Recreational and cultural services (50%) Restaurants and hotels Catering services Accommodation services

Excluded from the scope of VAT in 2010 CP040: CP044: CP060: CP062: CP063: CP100: CP101: CP102: CP103: CP105: CP120: CP125: CP126:

Housing, water, electricity, gas and other fuels Water supply and miscellaneous(50%) Health Out-patient services (50%) Hospital services (50%) Education Pre-primary and primary education Secondary education Post-secondary non-tertiary education Education not defined by level Miscellaneous goods and services Insurance Financial services n. e. c.

For exposition purposes, we only report the evolution of the 1-digit categories, e.g., CP020, for goods and services subject to the standard rate in 2010. All 2-digit categories that do not appear in the other sections, e.g., CP041, are subject to the default tax rates of the associated 1-digit category.

46

Table B2. Tax compliance fluctuations across regions.

VAT Compliance Firm size (assets, M euros) [.174]

(1)

(2)

(3)

(4)

(5)

-1.341∗∗∗ (.332)

-1.256∗∗∗ (.211)

-1.388∗∗∗ (.211)

-1.204∗∗∗ (.188)

-0.793 (1.79)

-1.484∗∗∗ (.174)

-1.425∗∗∗ (.169)

-1.733∗∗∗ (.174)

-1.056 (.960)

Effective VAT change [.149] Firm size × VAT change

Controls (GDP growth) Controls (sectoral growth) Extended controls Observations Adjusted R-squared

-4.642 (.6.85)

50 0.237

50 0.692

Yes

Yes Yes

50 0.716

50 0.856

Yes Yes Yes 50 0.833

Significantly different than zero at ∗ 90% confidence, ∗∗ 95% confidence, ∗∗∗ 99% confidence. Standard errors between parentheses are robust. The averages over the sample are shown between brackets. Extended controls include the interactions between firm size and sectoral growth for 10 sectors.

Table B3. Austerity in Greece – evolution of leverage by firm size (panel).

Leverage Post-Recession (assets less than 2M)

(1) -.0218∗∗∗ (.0011)

(2) -.0150∗∗∗ (.0018)

Post-Recession (assets between 2M and 20M)

-.0494∗∗∗ (.0018)

-.0371∗∗∗ (.0018)

Post-Recession (assets between 20M and 50M)

-.0005 (.0038)

.0004 (.0038)

Observations Firms Fixed effects Sectoral trend

187,705 37,540 Yes No

187,705 37,540 Yes Yes

Significantly different than zero at ∗ 90% confidence, ∗∗ 95% confidence, ∗∗∗ 99% confidence. Robust standard errors are reported between parentheses.

47

Figures Appendix Figure B1. VAT compliance and excise compliance in Greece (2007-2012).

(a) VAT compliance.

(b) Excise compliance.

Note: Source Hellastat, 2009, 2011. The left (resp. right) panel represents the ratio of realized tax revenues to expected tax revenues, given the consumption of goods subject to different VAT rates. We report the VAT effective rate (resp. the excise rate) on the right axis and the associated VAT (resp. excise) compliance over the period 2007-2012 in Greece.

Figure B2. Response of VAT compliance to the 2010 VAT reform and the role of tax pressure and firm size.

(a) Change in effective VAT rate

(b) Firm size

Note: In the left panel (resp. right panel), we report the correlation between the evolution of VAT compliance, i.e., the (log) difference of regional VAT compliances after and before the 2010 tax reform, and the percentage change in effective VAT rates (resp. the average regional firm size). In both figures, the circle size illustrate the share of national activity for each region. For readibility purposes, we omit the Attic region (very high share of total activity, very high tax compliance and large negative response to the 2010 reform).

48

Figure B3. Empirical production function.

(a) Polynomial estimates

(b) Density

Note: These figures represent the polynomial estimates for the elasticity of sales to firm endowment using the whole sample of firms (approximately 30’000 firms per year) and controlling for firm and industry×year fixed effects. For both figures, the axes are on a logarithmic scale.

Figure B4. Leverage as a function of firm size before and after the 2010 tax reform for the subsamples of regions with high/low tax compliance response.

(a) High response of tax compliance

(b) Low response of tax compliance

Note: Source Hellastat, 2009, 2011. This graph displays the leverage by firm size (total assets) before (2009) and after (2011) the austerity plan in regions with above-median response in tax compliance (left panel) and below-median response in tax compliance (right panel).

49

Figure B5. VAT compliance and leverage.

Note: We report the correlation between the evolution of VAT compliance, i.e., the (log) difference of regional VAT compliances after and before the 2010 tax reform, and the evolution of leverage, i.e., the (log) difference of leverage after and before the 2010 tax reform. In both figure, the circle size illustrate the share of national activity for each region. For readibility purposes, we omit the Attic region.

50